How to process a very large file

mattca · 11-22-2013, 02:22 PM

Hi all, I have a large dataset in a single json file that I need to process somehow. This file is bigger than my available RAM (although smaller than total RAM + swap space). I had written a simple python script to process it that I had tested with smaller datasets, but obviously I run out of memory when I try to open the file I actually need to process.

What's the best way to proceed? I'm not limited to using python, I can use whatever tool suits the job.

If what I need to do with it matters, I basically need to turn something like this:

Code:

[
    {"fruit": "apple", "vegetable": "pea"},
    {"fruit": "banana", "vegetable": "carrot"},
    {"fruit": "grapefruit", "vegetable": "pea"},
    {"fruit": "apple", "vegetable": "celery"}
]

into something like this:

Code:

{
    "pea": ["apple", "grapefruit"],
    "carrot": ["banana"],
    "celery": ["apple"]
}

Basically, I have a list of rows, and need to create an object indexed by "vegetable", which contains a list of the "fruits" associated with that vegetable.

Does that make sense?

Anyway, what's the best way to approach this? The file is 22G, and I have 8G RAM and 16G swap.

Edit: if it helps, here's the python script I wrote

Code:

#!/usr/bin/python

import sys
import json

if len(sys.argv) != 3:
    print "Usage:"
    print "python process-results.py <input file> <output file>"
    sys.exit()

inputfilename = sys.argv[1]
outputfilename = sys.argv[2]

with open(inputfilename) as infile:
    items = json.load(infile)
    filtered_items = {}

    for item in items:
        vegetable = item["vegetable"][0]

        if vegetable not in filtered_items:
            filtered_items[vegetable] = []

        filtered_items[vegetable].append(item["fruit"])

    infile.close()

with open(outputfilename, "w") as outfile:
    json.dump(filtered_items, outfile, indent=4)
    outfile.close()

danielbmartin · 11-22-2013, 02:31 PM

Consider using the split command to break your huge input file into manageable subsets, and process each individually.

Daniel B. Martin

syg00 · 11-22-2013, 02:49 PM

Use ijson instead ?.

Guttorm · 11-22-2013, 02:58 PM

Split would have to be merged, so it would need a bit more coding. Why does it read it all into an array and then iterate? If you read the file line by line as text, remove the last comma, it's json. Parse each line and add to filtered_items, like you have code for already.

NevemTeve · 11-23-2013, 02:36 AM

sort(1) is your friends

Input:

Code:

[
    {"fruit": "apple", "vegetable": "pea"},
    {"fruit": "banana", "vegetable": "carrot"},
    {"fruit": "grapefruit", "vegetable": "pea"},
    {"fruit": "apple", "vegetable": "celery"}
]

output of step#1 (python, perl, awk, bash etc):

Code:

"pea" "apple"
"carrot" "banana"
"pea" "grapefruit"
"celery" "apple"

output of sort:

Code:

"carrot" "banana"
"celery" "apple"
"pea" "apple"
"pea" "grapefruit"

output of step#3 (python, perl, awk, bash etc):

Code:

"carrot": "banana"
"celery": "apple"
"pea": "apple", "grapefruit"

dugan · 11-25-2013, 11:37 AM

Quote:

Originally Posted by mattca

Anyway, what's the best way to approach this? The file is 22G, and I have 8G RAM and 16G swap.

Honestly, the best way to approach this is to find a computer with 32GB RAM to run it on. It can be a physical one, or you can rent CPU time on a cloud.

Okay...

If that's not an option, then I would attempt the following: Read the file line-by-line once to determine the keys ("carrot", "celery", "pea" in the example), create a database with a column header for each key, reread the input file line by line and update the database for each line, and then *separately* write the contents of each database column as a key in the output file. That way, only the contents of one column/key will ever need to be in memory at a time.

chrism01 · 11-29-2013, 05:25 AM

I'd consider a tied hash in Perl; basically means you get ability to x-ref keys+values as you go line-by-line, but the hash is actually stored on disk, not in mem

pan64 · 11-29-2013, 05:48 AM

I would try to read file only line by line, replace strings with numbers (or sorter strings), so 1 is apple, 2 is grapefruit ...
You will only need to create a simple mapping table and handle those numbers. You can use arrays to store list of numbers (vegetables) and another array (of fruits) to store those arrays. Sorting can also help a lot as it was mentioned. Just remember, you need not keep all the file in memory.